Causal inference at Ghent University

Ghent University has a long tradition of research in causal inference since the mid 90's, initiated by my former PhD advisor Prof. Els Goetghebeur. Today, our vibrant causal inference community comprises over 20 statisticians dedicated to advancing this field.

Currently, I am honored to collaborate closely with Prof. Oliver Dukes, alongside postdoctoral fellows Heidelinde Dehaene, Johan Steen, and Kelly Van Lancker. My team is further enriched by the contributions of doctoral students that I (co-)supervise: Muluneh Alene, Eline Anslot, Georgi Baklicharov, Alexander Decruyenaere, Edoardo Gervasoni and Wout Waterschoot. I also co-supervise Katherine Holdsworth and Matthew Pryce from the London School of Hygiene and Tropical Medicine.

In addition to our core team, this academic year, we are delighted to welcome visiting PhD students Pan Zhao from Montpellier and Zehao Su from Copenhagen.

About me

I am a biostatistician with over 20 years of experience in the development of statistical methods for causal inference. I have spent most of my career at Ghent University, except for postdoctoral training under the guidance of Andrea Rotnitzky and James Robins at the Department of Biostatistics of the Harvard School of Public Health, and a part-time professorship as Professor of Statistical Methodology at the Department of Medical Statistics at the London School of Hygiene and Tropical Medicine from 2017 to 2021.

I have had the privilege of serving as Co-Editor of Biometrics, the flagship journal of the International Biometric Society. Furthermore, I have contributed as an Associate Editor for several prestigious journals including the American Journal of Epidemiology, Biometrics, Biostatistics, Epidemiology, Epidemiologic Methods, the Journal of Causal Inference, and the Journal of the Royal Statistical Society–Series B.

On a personal note, I have 2 sons and find great joy in cooking, sourdough baking and long-distance running.

News

April 11, 2024 - I am deeply honored and humbled to be recipient of an advanced ERC award to spearhead research on Assumption-lean (Causal) Modeling and Estimation. Thank you so much to my former mentors and my long-time collaborators for invaluable guidance, inspiration and encouragement; to the ERC PE1 panelists and reviewers for their endorsement and confidence in this project; and to the exceptional causal inference team at Ghent University, which continually triggers my passion for methodological research driven by real-world challenges.

In an era where the focus on causal inference is increasingly turning away from modeling towards quantifying population-level intervention effects, there is a risk of oversimplifying causal queries and of neglecting the rich history and efficacy of statistical modeling techniques. This ERC project aims to bridge this gap by leveraging the flexibility and power of statistical models to accurately represent facets of the causal data-generating mechanism, integrating it with recent insights from debiased machine learning and causal inference. Besides laying foundations for a novel paradigm for statistical modeling, rooted in the read paper by Vansteelandt and Dukes (2022) in the Journal of The Royal Statistical Society - Series B , titled "Assumption-lean inference for generalized linear model parameters," this project seeks to enhance the robustness and efficiency of debiased machine learning methods.

In the near future, I will be announcing openings for two postdoctoral fellow positions and four doctoral student positions to join my research team in this exciting journey.

Research

My research primarily revolves around developing statistical methods to infer the causal effects of exposures, treatments, or interventions on specific endpoints. I advocate for minimalism in modeling assumptions, favoring semi-parametric methods and debiased machine learning techniques. My goal is to deliver methods that offer stability and robustness, making them accessible to non-experts.

I collaborate closely with critical care clinicians to analyze routinely collected health data. Additionally, I work with pharmaceutical statisticians at Johnson & Johnson, GSK Belgium, and Novo Nordisk to tackle statistical challenges in the evaluation of medicines and vaccines.

Assumption-lean Modelling

I strive to create a versatile and user-friendly data modeling framework that combines the strengths of statistical modeling and debiased machine learning. This framework aims to minimize bias, enhance interpretability, and provide honest inferences.

Causal Machine Learning

My focus lies in crafting data-adaptive estimation strategies for evaluating the effects of exposures on time-to-event and other endpoints. I moreover study how to improve the efficiency and robustness of causal machine learning techniques.

Causal Prediction

I develop orthogonal statistical learning strategies tailored for estimating covariate-specific treatment effects, exploiting infinite-dimensional targeted learning to improve performance.

Causal Mediation Analysis

I develop data analysis techniques aimed at investigating causal mechanism. This involves assessing the extent to which the effect of an exposure on an outcome is mediated by specific intermediate variables.

Covariate Adjustment and Intercurrent Events in RCTs

I develop techniques for improving the efficiency of RCT analyses by exploiting baseline and intermediate covariates. I moreover study how to deal with intercurrent events that pose challenges to the interpretation of RCT analyses.

Inference for Synthetic Data

With the SYNDARA team, I study the quality of statistical analyses based on synthetic data obtained from generative adversarial networks. I strive to propose methods that ensure valid inference based on synthetic data.

Papers and recent talks